Process for monitoring processing plants

• US 6,618,632 B1
• Filed: 04/24/2001
• Issued: 09/09/2003
• Est. Priority Date: 12/11/1997
• Status: Expired due to Fees

• Current Assignee: Siemens AG
• Original Assignee: Infineon Technologies AG
• Inventors: Wachtmeister, Frank, Federl, Peter, Pöppel, Gerhard
• Primary Examiner(s): Patel, Ramesh
• Assistant Examiner(s): PHAM, THOMAS K
• Application Number:
  US09/581,665
• Time in Patent Office: 868 Days
• Field of Search: 700/14, 700/47, 700/55, 700/121, 700/159, 700/110, 700/143
• US Class Current: 700/47
• CPC Class Codes: G05B 23/0221 Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks

Abstract

A method of monitoring and/or controlling machining plans having time-dependent machining parameters includes measuring desired time-dependent machining parameters as a measured curve, generating time-independent numerical values from the measured machining parameters, and entering the time-independent numerical values into a classifier distinguishing between normal states of the machining plant and abnormal states of the machining plant. Training vectors regarded as abnormal can be filtered out from training vectors available by determining a distance of each training vector from every other training vector with a selected measure of distance for filtering of the abnormal training vectors, the training vectors having components made of time-independent numerical values, and the classifier can be trained with training vectors.

Claims

1. A method of monitoring and/or controlling machining plants having time-dependent machining parameters, which comprises:

measuring desired time-dependent machining parameters of a machining plant as a measured curve;

generating time-independent numerical values from the measured machining parameters;

entering the time-independent numerical values into a classifier distinguishing between normal states of the machining plant and abnormal states of the machining plant;

filtering out training vectors regarded as abnormal from training vectors available by determining a distance of each training vector from every other training vector with a selected measure of distance for filtering of the abnormal training vectors, the training vectors having components made of time-independent numerical values; and

training the classifier with training vectors.

2. The method according to claim 1, which further comprises providing a neural network as the classifier.

3. The method according to claim 1, which further comprises using one of a group consisting of a hypersquare and a hypersphere with a radius λ

_j as a domain around a prototype.

4. The method according to claim 1, which further comprises satisfying requirements of universality, similitude, robustness, and normalization with the time-independent numerical values.

5. The method according to claim 1, which further comprises filtering out training vectors for which a distance value K of an ordered sequence of distances in which a prescribed percentage of all the vectors have smaller distances from a training vector in question exceeds a prescribed threshold value.

6. The method according to claim 1, which further comprises using weighted measures of distance as measures of distance for determining a distance of each training vector from every other training vector.

7. The method according to claim 2, which further comprises providing an RCE network as the classifier.

8. The method according to claim 4, which further comprises approximating time-dependent measured curves of a desired machining parameter by a linear combination of polynomials, and using coefficients c_i of the linear combination as time-dependent numerical values.

9. The method according to claim 4, which further comprises using, as numerical values, distances of the measured curve with respect to prescribed comparison curves.

10. The method according to claim 4, which further comprises calculating an average measured curve from the measured curves of a machining parameter and using deviations with respect to the average measured curve as numerical values.

11. The method according to claim 4, which further comprises using, as numerical values, characteristic variables for characterizing the measured curves conceived as bit maps.

12. The method according to claim 4, which further comprises:

subdividing the measured curve into portions; and

using, as numerical values, at least one of mean values and standard deviations from the portions.

13. The method according to claim 6, which further comprises weighting the measures of distance according to at least one of information content, information entropy, a function of the information content of the respective numerical value, and a function of the information entropy of the respective numerical value.

14. The method according to claim 8, which further comprises carrying out the approximating step by approximating time-dependent measured curves of a desired machining parameter by a linear combination of orthogonal polynomials.

15. A method of filtering vectors regarded as abnormal from a set of vectors, which comprises:

determining a distance of each vector from every other vector by a selected measure of distance; and

filtering out vectors for which a distance value K of ordered sequence of distances in which a prescribed percentage of all the vectors have smaller distances from a vector in question exceeds a prescribed threshold value.

16. The method according to claim 15, which further comprises using weighted measures of distance as measures of distance for determining the distance of each vector from every other vector.

17. The method according to claim 16, which further comprises weighting the measures of distance according to at least one of information content, information entropy, a function of the information content of the respective vector components, and a function of the information entropy of the respective vector components.